EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Flashinfer sample kernel (#617)

Browse Source
This commit is contained in:
Liangsheng Yin
2024-07-17 13:24:43 -07:00
committed by GitHub
parent 4efcc59d4f
commit 3de2f30a27
4 changed files with 17 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/bench_latency.py
View File

@@ -156,14 +156,14 @@ def extend(reqs, model_runner):
)
batch.prepare_for_extend(model_runner.model_config.vocab_size, None)
output = model_runner.forward(batch, ForwardMode.EXTEND)
next_token_ids, _ = batch.sample(output.next_token_logits)
next_token_ids = batch.sample(output.next_token_logits)
return next_token_ids, output.next_token_logits, batch
def decode(input_token_ids, batch, model_runner):
batch.prepare_for_decode(input_token_ids.cpu().numpy())
output = model_runner.forward(batch, ForwardMode.DECODE)
next_token_ids, _ = batch.sample(output.next_token_logits)
next_token_ids = batch.sample(output.next_token_logits)
return next_token_ids, output.next_token_logits